Skip to content

Signal Validation Service – Basics

Checks

Range Check

Assets typically have inherent technical boundaries which are predefined by design. These boundaries set limits for acceptable values of measurement data. Range Check provides the possibility to define an upper and a lower limit for acceptable values of measurement data. It detects, if measurement data is outside these boundaries.

Data Gap Analysis

Time series data can be incomplete for various reasons:

  • There may be time gaps in the data
  • A data point may contain a timestamp but no value
  • Data may arrive without or with inconsistent units
  • Data may use a wrong format.

Many data analytics algorithms cannot deal with such gaps. For example, rules require data without gaps. For correlation analysis of different signal channels, a mapping of data to distinct timestamps is required.

The Data Gap Analysis can report gaps and perform linear interpolation.

Alerts

Spike Alert

The Spike Alert applies a three-sigma range to detect outliers: The algorithm uses a sliding window with a user defined size. When there are sufficient values available for the size of the sliding window, the window moves forward each time a new data point arrives. In each step, the mean and the standard deviation (sigma) are calculated using all data points within the window.
If the deviation of the newest data point, which is the last point in the window, from the mean exceeds 3 times sigma, an alert is returned.

Spike Alert

Noise Alert

Measurement noise can be caused by unintentional inaccuracies in the measuring process. Using noisy data in data analytics tasks can lead to various problems, e.g., wrong statistics or bad models.

The Noise Alert algorithm uses a sliding window with a user defined size. When there are sufficient values available for the size of the sliding window, the window moves forward each time a new data point arrives. To detect noise, the absolute distance to mean (ADTM) and the standard deviation (sigma) are calculated in the sliding window.

An incoming signal is treated as noise, if the standard deviation (sigma) from ADTM exceeds a user defined threshold.

The following diagram shows an example of the data analyzed using Noise Alert. The green data points are data without noise, the turquoise data points are noisy data correctly detected by the Noise Alert. The red data points are False Positives, that are wrongly labeled as noise and occur only in the short transition period where the machine’s behavior changes from normal to noisy.

Noise Alert

Jump Alert

A jump, i.e., a sudden large change in a signal sequence, can be indicator of some serious fault in industrial processes. Therefore, a mechanism is needed to identify such behavior and trigger an alert.

The Jump Alert algorithm uses a sliding window with a user defined size. When there are sufficient values available for the size of the sliding window, the window moves forward each time a new data point arrives. To detect spikes, the absolute distance to mean (ADTM) and the standard deviation (sigma) are calculated in the sliding window.

An alert is triggered, if two out of three consecutive signal values exceed 2 times sigma.

Chart

Bias Alert

Signal values should not deviate too much from an expected pattern, that represents normal behavior of the assets in a certain environment. The pattern can be a constant or given by a linear or non-linear function, e.g., sinus or even more functions.

Bias Alert detects changes in patterns and trends, e.g., when signal values continuously exceed a linear deviation threshold.

Bias Alert uses two sliding windows with a same user defined size. When there are sufficient values available to fill the two sliding window without overlapping, the windows move forward each time a new data point arrives. In each step, the data points of each window are transformed into frequency space by using Fast Fourier Transform (FFT). Then, their cosine similarity is calculated in frequency space.

A bias alert is triggered, if the similarity does not reach a user defined threshold.

Any questions left?

Ask the community


Except where otherwise noted, content on this site is licensed under the MindSphere Development License Agreement.